mtdblock分析

#include
#include
#include
#include
#include
#include
#include
#include

#include
#include
#include


static struct mtdblk_dev {
    struct mtd_info *mtd;
    int count;
    struct mutex cache_mutex;
    unsigned char *cache_data;
    unsigned long cache_offset;
    unsigned int cache_size;
    enum { STATE_EMPTY, STATE_CLEAN, STATE_DIRTY } cache_state;
} *mtdblks[MAX_MTD_DEVICES];

static struct mutex mtdblks_lock;

/*
 * Cache stuff...
 *
 * Since typical flash erasable sectors are much larger than what Linux's
 * buffer cache can handle, we must implement read-modify-write on flash
 * sectors for each block write requests.  To avoid over-erasing flash sectors
 * and to speed things up, we locally cache a whole flash sector while it is
 * being written to until a different sector is required.
 */

static void erase_callback(struct erase_info *done)//擦除回调函数
{
    wait_queue_head_t *wait_q = (wait_queue_head_t *)done->priv;
    wake_up(wait_q);
}
//擦除写函数，mtd为待擦除的块设备，例如某个分区设备
static int erase_write (struct mtd_info *mtd, unsigned long pos,
            int len, const char *buf)
{
    struct erase_info erase;
    DECLARE_WAITQUEUE(wait, current);
    wait_queue_head_t wait_q;
    size_t retlen;
    int ret;

    /*
     * First, let's erase the flash block.
     */

    init_waitqueue_head(&wait_q);
    erase.mtd = mtd;
    erase.callback = erase_callback;
    erase.addr = pos;
    erase.len = len;
    erase.priv = (u_long)&wait_q;

    set_current_state(TASK_INTERRUPTIBLE);
    add_wait_queue(&wait_q, &wait);
//调用设备的擦除函数，例如对于分区设备是part_erase，如果主设备没有分区，则为//cfi_amdstd_erase_varsize，实际上分区设备最后都是调用了主设备的cfi_amdstd_erase_varsize
    ret = mtd->erase(mtd, &erase);
    if (ret) {
        set_current_state(TASK_RUNNING);
        remove_wait_queue(&wait_q, &wait);
        printk (KERN_WARNING "mtdblock: erase of region [0x%lx, 0x%x] "
                     "on \"%s\" failed\n",
            pos, len, mtd->name);
        return ret;
    }

    schedule();  /* Wait for erase to finish. */
    remove_wait_queue(&wait_q, &wait);

    /*
     * Next, writhe data to flash.
     */
//调用设备的些函数，对于分区设备是part_write
    ret = mtd->write(mtd, pos, len, &retlen, buf);
    if (ret)
        return ret;
    if (retlen != len)
        return -EIO;
    return 0;
}

//写缓存数据到FLASH
static int write_cached_data (struct mtdblk_dev *mtdblk)
{
    struct mtd_info *mtd = mtdblk->mtd;
    int ret;
//设备状态非脏
    if (mtdblk->cache_state != STATE_DIRTY)
        return 0;

    DEBUG(MTD_DEBUG_LEVEL2, "mtdblock: writing cached data for \"%s\" "
            "at 0x%lx, size 0x%x\n", mtd->name,
            mtdblk->cache_offset, mtdblk->cache_size);
//把cache数据写入flash
    ret = erase_write (mtd, mtdblk->cache_offset,
               mtdblk->cache_size, mtdblk->cache_data);
    if (ret)
        return ret;

    /*
     * Here we could argubly set the cache state to STATE_CLEAN.
     * However this could lead to inconsistency since we will not
     * be notified if this content is altered on the flash by other
     * means.  Let's declare it empty and leave buffering tasks to
     * the buffer cache instead.
     */
    mtdblk->cache_state = STATE_EMPTY;// 状态置为空
    return 0;
}


static int do_cached_write (struct mtdblk_dev *mtdblk, unsigned long pos,
                int len, const char *buf)
{
    struct mtd_info *mtd = mtdblk->mtd;
    unsigned int sect_size = mtdblk->cache_size;
    size_t retlen;
    int ret;

    DEBUG(MTD_DEBUG_LEVEL2, "mtdblock: write on \"%s\" at 0x%lx, size 0x%x\n",
        mtd->name, pos, len);

    if (!sect_size)
        return mtd->write(mtd, pos, len, &retlen, buf);

    while (len > 0) {
        unsigned long sect_start = (pos/sect_size)*sect_size;//计算写起始地址（块对齐）
        unsigned int offset = pos - sect_start;//块内起始偏移
        unsigned int size = sect_size - offset;//每次写的数据大小
        if( size > len )
            size = len;

        if (size == sect_size) {//如果写的大小等于块大小，直接erase_write擦除后写入
            /*
             * We are covering a whole sector.  Thus there is no
             * need to bother with the cache while it may still be
             * useful for other partial writes.
             */
            ret = erase_write (mtd, pos, size, buf);
            if (ret)
                return ret;
        } else {
            /* Partial sector: need to use the cache */

            if (mtdblk->cache_state == STATE_DIRTY &&
                mtdblk->cache_offset != sect_start) {//脏且起始偏移不等于块起始
                ret = write_cached_data(mtdblk);
                if (ret)
                    return ret;
            }

            if (mtdblk->cache_state == STATE_EMPTY ||
                mtdblk->cache_offset != sect_start) {{//空或者起始偏移不等于块起始
                /* fill the cache with the current sector */
                mtdblk->cache_state = STATE_EMPTY;
                ret = mtd->read(mtd, sect_start, sect_size,
                        &retlen, mtdblk->cache_data);//先读取扇区数据
                if (ret)
                    return ret;
                if (retlen != sect_size)
                    return -EIO;

                mtdblk->cache_offset = sect_start;//写偏移置为起始地址
                mtdblk->cache_size = sect_size;
                mtdblk->cache_state = STATE_CLEAN;
            }

            /* write data to our local cache */
            memcpy (mtdblk->cache_data + offset, buf, size);//把数据拷贝到cache
            mtdblk->cache_state = STATE_DIRTY;//状态为脏，下次写入
        }

        buf += size;
        pos += size;
        len -= size;
    }

    return 0;
}


static int do_cached_read (struct mtdblk_dev *mtdblk, unsigned long pos,
               int len, char *buf)
{
    struct mtd_info *mtd = mtdblk->mtd;
    unsigned int sect_size = mtdblk->cache_size;
    size_t retlen;
    int ret;

    DEBUG(MTD_DEBUG_LEVEL2, "mtdblock: read on \"%s\" at 0x%lx, size 0x%x\n",
            mtd->name, pos, len);

    if (!sect_size)
        return mtd->read(mtd, pos, len, &retlen, buf);

    while (len > 0) {
        unsigned long sect_start = (pos/sect_size)*sect_size;
        unsigned int offset = pos - sect_start;
        unsigned int size = sect_size - offset;
        if (size > len)
            size = len;

        /*
         * Check if the requested data is already cached
         * Read the requested amount of data from our internal cache if it
         * contains what we want, otherwise we read the data directly
         * from flash.
         */
        if (mtdblk->cache_state != STATE_EMPTY &&
            mtdblk->cache_offset == sect_start) {//如果状态不为空且起始偏移等于块起始地址
            memcpy (buf, mtdblk->cache_data + offset, size);
        } else {
            ret = mtd->read(mtd, pos, size, &retlen, buf);
            if (ret)
                return ret;
            if (retlen != size)
                return -EIO;
        }

        buf += size;
        pos += size;
        len -= size;
    }

    return 0;
}

static int mtdblock_readsect(struct mtd_blktrans_dev *dev,
                  unsigned long block, char *buf)
{
    struct mtdblk_dev *mtdblk = mtdblks[dev->devnum];
    return do_cached_read(mtdblk, block<<9, 512, buf);
}

static int mtdblock_writesect(struct mtd_blktrans_dev *dev,
                  unsigned long block, char *buf)
{
    struct mtdblk_dev *mtdblk = mtdblks[dev->devnum];
    if (unlikely(!mtdblk->cache_data && mtdblk->cache_size)) {
        mtdblk->cache_data = vmalloc(mtdblk->mtd->erasesize);
        if (!mtdblk->cache_data)
            return -EINTR;
        /* -EINTR is not really correct, but it is the best match
         * documented in man 2 write for all cases.  We could also
         * return -EAGAIN sometimes, but why bother?
         */
    }
    return do_cached_write(mtdblk, block<<9, 512, buf);
}

static int mtdblock_open(struct mtd_blktrans_dev *mbd)
{
    struct mtdblk_dev *mtdblk;
    struct mtd_info *mtd = mbd->mtd;
    int dev = mbd->devnum;

    DEBUG(MTD_DEBUG_LEVEL1,"mtdblock_open\n");

    mutex_lock(&mtdblks_lock);
    if (mtdblks[dev]) {
        mtdblks[dev]->count++;
        mutex_unlock(&mtdblks_lock);
        return 0;
    }

    /* OK, it's not open. Create cache info for it */
    mtdblk = kzalloc(sizeof(struct mtdblk_dev), GFP_KERNEL);
    if (!mtdblk) {
        mutex_unlock(&mtdblks_lock);
        return -ENOMEM;
    }

    mtdblk->count = 1;
    mtdblk->mtd = mtd;

    mutex_init(&mtdblk->cache_mutex);
    mtdblk->cache_state = STATE_EMPTY;
    if ( !(mtdblk->mtd->flags & MTD_NO_ERASE) && mtdblk->mtd->erasesize) {
        mtdblk->cache_size = mtdblk->mtd->erasesize;
        mtdblk->cache_data = NULL;
    }

    mtdblks[dev] = mtdblk;
    mutex_unlock(&mtdblks_lock);

    DEBUG(MTD_DEBUG_LEVEL1, "ok\n");

    return 0;
}

static int mtdblock_release(struct mtd_blktrans_dev *mbd)
{
    int dev = mbd->devnum;
    struct mtdblk_dev *mtdblk = mtdblks[dev];

       DEBUG(MTD_DEBUG_LEVEL1, "mtdblock_release\n");

    mutex_lock(&mtdblks_lock);

    mutex_lock(&mtdblk->cache_mutex);
    write_cached_data(mtdblk);
    mutex_unlock(&mtdblk->cache_mutex);

    if (!--mtdblk->count) {
        /* It was the last usage. Free the device */
        mtdblks[dev] = NULL;
        if (mtdblk->mtd->sync)
            mtdblk->mtd->sync(mtdblk->mtd);
        vfree(mtdblk->cache_data);
        kfree(mtdblk);
    }

    mutex_unlock(&mtdblks_lock);

    DEBUG(MTD_DEBUG_LEVEL1, "ok\n");

    return 0;
}

static int mtdblock_flush(struct mtd_blktrans_dev *dev)
{
    struct mtdblk_dev *mtdblk = mtdblks[dev->devnum];

    mutex_lock(&mtdblk->cache_mutex);
    write_cached_data(mtdblk);
    mutex_unlock(&mtdblk->cache_mutex);

    if (mtdblk->mtd->sync)
        mtdblk->mtd->sync(mtdblk->mtd);
    return 0;
}

static void mtdblock_add_mtd(struct mtd_blktrans_ops *tr, struct mtd_info *mtd)
{
    struct mtd_blktrans_dev *dev = kzalloc(sizeof(*dev), GFP_KERNEL);

    if (!dev)
        return;

    dev->mtd = mtd;
    dev->devnum = mtd->index;

    dev->size = mtd->size >> 9;
    dev->tr = tr;

    if (!(mtd->flags & MTD_WRITEABLE))
        dev->readonly = 1;

    add_mtd_blktrans_dev(dev);
}

static void mtdblock_remove_dev(struct mtd_blktrans_dev *dev)
{
    del_mtd_blktrans_dev(dev);
    kfree(dev);
}

static struct mtd_blktrans_ops mtdblock_tr = {
    .name        = "mtdblock",
    .major        = 31,
    .part_bits    = 0,
    .blksize     = 512,
    .open        = mtdblock_open,
    .flush        = mtdblock_flush,
    .release    = mtdblock_release,
    .readsect    = mtdblock_readsect,
    .writesect    = mtdblock_writesect,
    .add_mtd    = mtdblock_add_mtd,
    .remove_dev    = mtdblock_remove_dev,
    .owner        = THIS_MODULE,
};

static int __init init_mtdblock(void)
{
    mutex_init(&mtdblks_lock);

    return register_mtd_blktrans(&mtdblock_tr);
}

static void __exit cleanup_mtdblock(void)
{
    deregister_mtd_blktrans(&mtdblock_tr);
}

module_init(init_mtdblock);
module_exit(cleanup_mtdblock);


MODULE_LICENSE("GPL");
MODULE_AUTHOR("Nicolas Pitre et al.");
MODULE_DESCRIPTION("Caching read/erase/writeback block device emulation access to MTD devices");